KR101704096B1 - Process for performing log-likelihood-ratio clipping in a soft-decision near-ml detector, and detector for doing the same - Google Patents

Abstract

The method calculates a log-likelihood-ratio (LLR) at the detector of the wireless communication receiver, and the LLR is used by the channel decoder. The method comprises the steps of receiving a signal from a telecom front end, the signal corresponding to data belonging to a finite set of constellation symbols, each constellation symbol having a lattice constant Calculating a limited set of distances representing a Euclidean distance between a received signal that can be multiplied by the channel and a finite set of predetermined constellation symbols, arranged in a stellar manner, Deriving a soft decision or an LLR set from a limited set of said distances under the condition of the SNR and bit index and completing the derived LLR by a clipping value read from a lookup table that is simultaneously addressed by the values of the SNR and bit index. The present invention is particularly suitable for embodying an OFDM receiver for a MIMO system.

Description

PROCESS FOR PERFORMING LOG-LIKELIHOOD-RATIO CLIPPING IN A SOFT-DECISION NEAR-ML DETECTOR, AND DETECTOR FOR DOING THE SAME.

Field of the Invention [0002] The present invention relates to the field of wireless communications, and more particularly, to a method for performing LLR computation in a soft decision near-maximum likelihood detector, and a receiver thereof.

Nowadays, wireless communications are becoming more and more popular and wireless networks are becoming more and more popular as more powerful modulation techniques such as M-ary amplitude modulation, or M-ary Quadrature Amplitude Modulation (QAM) .

QAM provides a constellation (with different combinations of phase and amplitude) of the M number of modulation values, and each constellation point (symbol) represents a plurality of information bits. The number of bits represented by each symbol in the M-ary QAM system is equal to log ₂ M. Different M-ary QAM constellations are widely spread from robust 4QAM to high-speed 64QAM constellations.

QAM techniques may be advantageously combined with more recent approaches such as OFDM (Orthogonal Frequency Division Multiplex) as well as Multiple In Multiple Out (MIMO) techniques.

In a MIMO system that includes M transmit antennas and N receive antennas, the receiver must process the M transmitted symbol sets from the N observed signal sets (e.g., in OFDM), and the signals are channel- and noise- It can be damaged by the characteristics. The role of the detector is to select between the received data and all possible transmitted symbol vectors based on the estimated channel. As is known by those skilled in the art, the detector that always returns the optimal solution is the so-called ML detector (Maximum Likelihood), but it is known to be extremely complex to implement.

However, this optimal maximum likelihood detector can be efficiently approximated by using several techniques, such as sphere decoding, lattice reduction, or a combination of the two, often referred to as soft quasi-quasi-maximal near-ML techniques . The soft decision sub-ML provides accurate Log-Likelihood-Ratio (LLR) calculations for most of the bits that make up the channel and MIMO encoded transmission sequence.

Generally speaking, the quasi ML detector generates a list of distances for a given bit to be used in the second step for the calculation of the LLR, as is basically illustrated in FIG.

In the case of a maximum log approximation, the above-described distance list is approximated by only two distances calculated between the received demodulated symbol with each bit value of 1 and 0 and the point of the hypothetical constellation symbol.

The LLR estimate for the k ^{- th} bits may be calculated according to the following formula, which is well known to those skilled in the art.

d ² _{1min , k} is the minimum distance between the received demodulation symbol and the QAM constellation point with a particular bit equal to 1, and d ² _{0min , k} is the distance between the received demodulation symbol and the QAM constellation point . Also,? ² represents a noise variation.

To reduce its complexity (with respect to the ML detector), the quasi-ML detector performs the calculation of the LLR only for the limited number of bits belonging to the sequence.

For other bits belonging to a sequence, the LLR is not explicitly computed and it is normally set to a predefined value. This operation is commonly referred to as LLR clipping, as specified in the following two references.

[1] B.M. Hochwald and S. ten Brink. &Quot; Achieving near-capacity on a multiple-antenna channel ". Communications, IEEE Transactions on, vol. 51, p. 389-399, Mar. 2003.

[2] Y. de Jong and T. Willink, "Iterative Tree Search Detection for MIMO wireless systems", Communications, IEEE Transactions on, vol. 53, p. 930-935, June 2005.

However, the choice of clipping level has a strong impact on system performance.

Generally speaking, if the selected clipping level is too high, this will interfere with the correction of the decision error that occurs at some bit positions and degrade performance. Conversely, if the selected clipping level is too low, this limits transmission information at the detector output and also reduces performance.

GA 30332-0250 papers due to David L. Milliner of Electrical and Computer Engineering, University of Technology, Atlanta, Georgia "Channel State Information Based Clipping in LLR List MIMO Detection "discloses an improved mechanism for using clipping levels based on knowledge of SNR (Signal to Noise Ratio). The result shows that the FLC (Fixed LLR Clipping Level) set at +3 and -3 according to the bit code Although this prior art has been shown to be significantly improved with respect to FLC, such a technique relies on the presence of a Gaussian signal that does not correspond to the reality of the discrete QAM constellation.

Therefore, there is a desire for a more appropriate mechanism.

It is an object of the present invention to provide a new method of calculating the LLR in a coded MIMO system that considers the multilevel bit mapping nature of quadrature amplitude modulation.

It is a further object of the present invention to provide an improved method of calculating the LLR in a coded MIMO system, in order to provide near-optimal per-bit detection performance while limiting the increase in complexity.

It is yet another object of the present invention to provide a method of calculating a log likelihood ratio in a MIMO soft decision quasi-ML detector.

Yet another object of the present invention is to provide a semi-ML of a sphere decoder which is designed to generate LLRs by applying bit-position and SNR knowledge and clipping values.

These and other objects of the present invention are achieved by a method for calculating a soft decision value or a log likelihood-ratio (LLR) in a detector of a wireless communication receiver, the method comprising: receiving a signal from a telecom front end; Each constellation symbol corresponding to data belonging to a finite set of constellation symbols, each constellation symbol being arranged in a lattice constellation which is impaired by additional noise and also by a multiplication channel,

Calculating a limited set of distances representing the Euclidean distance between the received signal that can be multiplied by the channel and a finite set of predetermined constellation symbols;

Deriving a set of soft decision or LLR (Log Likelihood Ratio) from the limited set of distances under the constraint of the length of the limited list of distances,

And completing the derived LLR by a clipping value read from a lookup table that is simultaneously addressed by the SNR and the value of the bit index for a distance not included in the limited distance list.

In one embodiment, there is a predetermined length of the limited distance list where the correct LLR is calculated by the detector. Then, a predetermined distance set is calculated according to the length.

The method then proceeds to the calculation of the LLR for a given set of distances and then completes such calculation by reading the lookup table which is addressed simultaneously by the value of the SNR and the bit index, and possibly by the index value.

In one embodiment, the constellation is a quadrature amplitude constellation such as 4QAM, 16QAM or 64QAM. The channel decoder may be a turbo decoder or a Viterbi decoder.

The present invention also provides a receiver for a wireless communication system including a detector,

Means for receiving a signal from a telecom front end, the signal corresponding to data belonging to a finite set of constellation symbols, each constellation symbol having a quadrature amplitude coefficient < RTI ID = 0.0 > Deployed as a < RTI ID = 0.0 >

Means for calculating a limited set of distances representing the Euclidean distance between a received signal that may be corrupted by the channel and a finite set of predetermined constellation symbols;

Means for deriving a set of soft decision or LLR (Log Likelihood Ratio) from the limited set of distances under the constraint of the length of the limited distance list,

And means for completing the derived LLR by a clipping value read from a lookup table that is simultaneously addressed by the SNR and bit index values for distances not included in the limited distance list.

The present invention is particularly adapted to embody a soft decision quasi-ML detector that may be used in a MIMO wireless communication system.

Other features of one or more embodiments of the invention will be best understood by reference to the following detailed description when read together with the accompanying drawings.
Figure 1 illustrates a prior art detector providing LLR computation.
FIG. 2 illustrates an example of a 16-QAM case indicating that there are two levels, including a MSB (Most Significant Bit) and an LSB (Least Significant Bit), for each signal point.
Figure 3 illustrates the LLR distribution of a full ML decoder and shows that the LLR size and distribution are dependent on SNR and bit level.
Figure 4 illustrates the distribution of LLRs normalized by the SNR of the described embodiment. It may further be noted that the LLRs are distributed differently within the same bit level depending on whether positive or negative are present. In the figure, it is clear that the defined LLR for the LSB is capped at an upper limit.
Figure 5 illustrates the maximum LLR value as a function of SNR.
6 illustrates a block diagram of one embodiment of a receiver in accordance with the present invention.
7A to 7D illustrate the BER performance of LSB ⁺ , LSB ^- , MSB ⁺ and MSB ^- , respectively.
Figure 8 illustrates a BER plot as a function of SNR for each of the 38% clipping values.
Figure 9 illustrates the method of the present invention.

Hereinafter, a method of augmenting the LLR calculation by an improved clipping selection mechanism will be described.

The present invention is particularly suitable for MIMO soft decision quasi-ML detectors that can be used for OFDM (Orthogonal Frequency Division Multiplexing).

However, it should be apparent that the methods and mechanisms described below are applicable to any other kind of modulation.

For clarity, examples of 16 QAM (Quadrature Amplitude Modulation) are contemplated, while those skilled in the art will simply apply the principles based on the following method to some other kind of modulation, such as 64QAM.

As shown in FIG. 2, 16QAM is based on a constellation plane defined by a horizontal axis representing the real part and a vertical axis representing the imaginary part. Thus, the constellation plane is divided into four quadrants, each quadrant containing four symbols.

Each symbol in the 16QAM system corresponds to log ₂ 16 or 4 bits, including LSB (Least Significant Bit) and MSB (Most Significant Bit), which can be distributed differently according to specific requirements. Figure 2 shows one exemplary mapping configuration between 4-bit and constellation points.

Figure 2 shows that different bits constituting one symbol represent "protection" or immunity to noise and size impairment. In particular, the MSB bit is protected by a higher Euclidean distance than the LSB since the receiver must switch from one quadrant to the other quadrant between the four quadrants in order to mistake one MSB bit. Conversely, for LSBs, illusions can occur in the same quadrant.

In the particular case of 16QAM, it is confirmed that the MSB and LSB are shared between the two groups, with each group being associated with one specific level of immunity or protection.

In the case of 64QAM, the different bits are distributed in three different groups representing three distinct levels of immunity or protection.

To take account of the different levels of immunity between the MSB and the LSB (in the exemplary example of 16QAM of FIG. 2), a mechanism is provided for setting the LLR to be provided to the channel decoder.

Indeed, the present invention provides a novel and effective method for considering the multi-level bitmapping nature of QAM, thereby providing a known mechanism and especially a conventional FLC (Fixed Log-likelihood-ratio Clipping) mechanism, or even a so-called SNR Log- -ratio Clipping.

Such a method is illustrated in FIG. 9, which illustrates the following steps used to calculate the LLR in the decoder.

- receiving a signal from the telecom front end (step 100), the signal corresponding to data belonging to a finite set of constellation symbols, each constellation symbol being impaired by additional noise and also by a multiplication channel Placed in a grid constellation.

Calculating a set of distances representing a Euclidean distance between a received signal that can be multiplied by the channel and a finite set of predetermined constellation symbols (step 200).

Deriving a set of soft decision or log likelihood ratio (LLR) from the limited set of distances under the constraint of the length of the limited distance list (step 300).

- completing the derived LLR by a clipping value read from a lookup table that is simultaneously addressed by the values of SNR and bit index (step 400).

More specifically, the specific length at which the correct LLR is used to determine the distance set or the limit distance list calculated by the detector is determined.

However, for other distances that are not included in the restricted list to limit the complexity of the method, the clipping value read from the lookup table which is simultaneously addressed by at least two values, i.e. the SNR and bit index, will then be returned.

In one embodiment, the constellation is a quadrature amplitude constellation such as 4QAM, 16QAM or 64QAM. The channel decoder may be a turbo decoder or a Viterbi decoder.

In one particular embodiment, the look-up table is specified by three parameters:

- SNR

- bit index

The order of the modulation

Figure 6 illustrates one simplified block diagram of one embodiment of a soft decision quasi-ML detector in accordance with the present invention.

The architecture includes one quasi-ML detector 61, e.g., a Sphere Decoder (SD) quasi-ML detector, which includes means for generating a list of distances for a given bit to be used in the second stage for the calculation of the LLR.

In the case of a maximum log approximation, the candidate list is constructed solely by the two distances calculated between the received signal point that signals both 1 and 0 to the considered bits and the closest symbol. The LLR for the k ^{- th} bits may be calculated by the LLR calculation block 62 according to any formula known to those skilled in the art, such as the conventional formula of the maximum log approximation provided as follows:

Where σ ² , d ² _{1min , k} and d ² _{Omin , k} represent the noise variance and the minimum Euclidean distance for the k ^{- th} transmitted bits of 1 and 0, respectively.

For a particular occurrence where the LLR is not exactly calculated (the detector 61 is assumed to be a semi-ML detector and not an ML detector), the LLR clipping block 63 is involved in determining one appropriate LLR clipping level to be considered.

In one embodiment, the LLR clipping level is extracted from the lookup table addressed by the following parameters: SNR, bit index, and modulation order.

This mechanism indicates that the sensitivity difference between the MSB and the LSB is very important and thus becomes very advantageous.

Figure 3 more specifically illustrates the LLR distribution of a complete ML decoder for six consecutive values of SNR: 8 dB, 9.5 dB, 11.0 dB, 12.0 dB, 13.5 dB, and 15 dB. The figure clearly shows that the LLR size and distribution depend on the SNR and also the bit level.

For FIG. 4, the LLR distribution is illustrated as soon as it is normalized by the SNR. It may further be noted that the LLRs are distributed differently within the same bit level depending on whether a positive or negative exists. In the figure, it is clear that the positive LLR for the LSB is capped at an upper limit.

Regarding the positive LSB and the MSB, the LLR still depends on the SNR. This point is more clearly illustrated in FIG. 5, where the maximum LLR at the SD sub-ML output is plotted as a function of the SNR for the different bit positions, MSB ^- , MSB ⁺ , LSB ^- and LSB ⁺ .

The positive LLR clipping value for the LSB denoted by LClip _LSB ⁺ is constant over the SNR range as previously mentioned. This differs from the case of the negative LLR clipping value for the LSB denoted by LClip _LSB ^- , depending on the SNR and on the number of clipped bits. Thus, LClip _LSB ^- ≠ LClip _LSB ⁺ .

The negative and positive LLR clipping values for the MSBs denoted LClip _MSB ^- and LClip _MSB ⁺ also depend on the SNR and on the number of clipped values.

As indicated by the simulation results, the LLR values are distributed differently according to the bit index and the SNR.

Thus, it can be seen that the LLR clipping mechanism can be significantly improved by exploiting this additional knowledge, resulting in low distortion and approximated LLR.

Figures 7a to 7d illustrate the BER (Bit Error Rate) obtained according to the three sizes of the distance set in the assumption of 10 dB SNR, which is a function of the possible clipping values and is 10 < ^-1 > And the BLER (BLock Error Rate).

For clarity and simplicity, all clipping values are individually selected, but this should be considered only as an example.

Simulation results are obtained for the different ratios of clipped LLR - 38%, 73% and 100%, respectively. This ratio is the ratio of the clipped value from the soft ML output. In the case of a ratio of p: p% of ML, the LLR is clipped to the test value. 100-p% Other LLR values are not modified.

All simulation results are compared with the ML BER used as a reference and a constant +/- 3 clipping value, which is presented as a highly efficient empirical result. See, for example, reference [2] above.

When the clipping value is separately considered a, BER performance is LSB ^+, LSB ^- are illustrated respectively in Figures 7a to 7d for the ^{-, +} MSB and MSB.

Figure 7A illustrates 38%, 73% and 100% of clipped values for a given SNR of 10dB and BER as a function of ML and +/- 3 clipping value references and test LSB ⁺ clipping values, respectively.

Figure 7b illustrates 38%, 73%, and 100% of the clipped values for a given SNR of 10 dB, and BER as a function of ML and +/- 3 clipping value references and test LSB ^- clipping values, respectively.

Figure 7C illustrates BER as a function of 38%, 73% and 100% of clipped values, and ML and +/- 3 clipping value references and test MSB ⁺ clipping values, respectively, for a given SNR of 10 dB.

Figure 7d illustrates 38%, 73%, and 100% of clipped values for a given SNR of 10dB, and BER as a function of ML and +/- 3 clipping value references and test MSB ^- clipping values, respectively.

It is found that the optimal clipping value for the positive LSB is less.

LClip _LSB ⁺ = + 0.20.

This optimal value is constant over the SNR range and is independent of the number of clipped bits. This different bit positions depending on the number of bits depends on the SNR and clipping LClip _LSB ^-, LClip _MSB ^-, not the case for LCIip _MSB ^+. In addition, it should be noted that the absolute clipping value for the MSB is the same as:

LClip _MSB ^- = -LClip _MSB ⁺

The BER performance verifies the efficiency with respect to the performance of this optimal clipping value according to bit position and SNR as shown in Fig.

As shown in FIG. 8, the proposed technique outperforms the +/- 3 clipping suggested by reference [2] by 0.22 dB.

The proposed technique allows low-distortion approximate LLR calculations at the output of a soft-decision quasi-ML detector. Moreover, such approximations are suitably applied according to the actual bitmapping indexing used by the QAM. As is known, any method provided so far utilizes this information to solve the problem of LLR clipping for the considered receivers, and such techniques provide significant performance improvements.

The invention has been shown to be particularly suitable for OFDM standards supporting the MIMO spatial plexing mode, e.g., IEEE 802.16, IEEE 802.11 and 3GPP LTE.

Claims (15)

A method of calculating a log likelihood ratio (LLR) used by a channel decoder or a soft decision value at a detector of a wireless communication receiver,
Receiving a signal from a telecom front end, the signal corresponding to data belonging to a finite set of constellation symbols, each constellation symbol corresponding to a set of constellation symbols, arranged in a lattice constellation that is damaged by a multiplicative channel,
Calculating a limited set of distances representing the Euclidean distance between the received signal and a finite set of predetermined constellation symbols that can be multiplied by the channel;
Deriving a set of soft decisions or LLRs from the limited set of distances under the constraint of the length of the limited distance list;
Completing the LLR by the clipping value read from the lookup table that is simultaneously addressed by the SNR and bit index values, and completing the LLR computation for distances not included in the restricted list of distances
Way.
The method according to claim 1,
Receiving a length of a limited distance list from which an accurate LLR can be calculated;
Calculating a predetermined distance set according to the limited distance list,
Generating a set of soft decisions or LLRs from the predetermined set of distances;
And completing the LLR with an LLR clipping value read from a lookup table that is simultaneously addressed by the SNR and the value of the bit index
Way.
The method according to claim 1,
The look-up table is also addressed by a third value indicating the modulation order
Way.
The method according to claim 1,
The constellation may be a quadrature amplitude constellation such as 4QAM, 16QAM or 64QAM.
Way.
The method according to claim 1,
The channel decoder may be a turbo decoder or a Viterbi decoder
Way.
The method according to claim 1,
The receiver may be an Orthogonal Frequency Division Multiplex (OFDM) receiver
Way.
The method according to claim 1,
The telecom front end may be an OFDM or CDMA communication system
Way.
And a detector for calculating a soft decision or LLR to be sent to the channel decoder,
Means for receiving a signal from a telecom front end, the signal corresponding to data belonging to a finite set of constellation symbols, each constellation symbol having a lattice constellation that is impaired by additional noise and also by a multiplication channel Deployed - and,
Means for calculating a limited set of distances representing the Euclidean distance between the received signal and a finite set of predetermined constellation symbols that can be multiplied by the channel;
Means for deriving a soft decision or LLR set from the limited set of distances under the constraint of the length of the limited distance list,
Means for completing the LLR by a clipping value read from a lookup table that is simultaneously addressed by a value of SNR and bit index and completing the LLR calculation for a distance not included in the restricted list of distances
receiving set.
9. The method of claim 8,
Means for receiving a length of a limited distance list from which an accurate LLR can be calculated,
Means for calculating a predetermined distance set according to the limited distance list,
Means for generating a set of soft decisions or LLRs from said predetermined set of distances,
Means for completing the LLR with an LLR clipping value read from a lookup table that is simultaneously addressed by the SNR and the value of the bit index
receiving set.
9. The method of claim 8,
The look-up table is also addressed by a third value indicating the modulation order
receiving set.
9. The method of claim 8,
The constellation may be a quadrature amplitude constellation such as 4QAM, 16QAM or 64QAM.
receiving set.
9. The method of claim 8,
The channel decoder may be a turbo decoder or a Viterbi decoder
receiving set.
9. The method of claim 8,
The receiver is an OFDM receiver
receiving set.
The method according to claim 1,
The telecom front end is a CDMA communication system
Way.
A receiver comprising a receiver as defined in any one of claims 8 to 13
Mobile phone.

Images